# Origin of increased noise in BF862/OP27 charge amplifier

For the amplification of piezoelectric sensor signals I am currently building a charge amplifier with a low noise JFET frontend utilizing an NXP BF862. The amplifier design is based on application notes from Linear Technology/Analog Devices (Design Note 254 and Design Note 308). My schematics and PCB layout are as follows:

My problem is that the measured noise is too high. For a capacitive test source (just a capacitor) of 264 pF I need an output referred noise floor below 100 nV/rtHz for frequencies between 800 Hz and 900 Hz. According to my LTspice simulation this should be achievable, the overall noise floor at 850 Hz is 52 nV/rtHz:

However, after producing the PCB several times with different layouts and measuring the noise floor at the amplifier output the result was always the same: Approx. 250 nV/rtHz at 850 Hz. Referred to the input that means approx. 9.5 nV/rtHz, which I think is much too high. I am doing these measurements in a faraday cage with a SR785 FFT analyzer whose input noise is well below 10 nV/rtHz. I changed the BF862 several times and I also increased the feedback resistance to up to 50 GOhm. But the noise floor always kept exactly the same. The amplifier was powered via two lead acid batteries. So, the simulated noise (52 nV/rtHz) is roughly a factor of five lower.

Does anybody have an idea what the problem could be, i.e. where the noise is coming from? Is it possible that the simulation is wrong? Or could it be that one of my components is not behaving like expected? For example I thought that maybe the feedback resistance (or the source/drain resistances) adds some kind of additional 1/f noise? I also checked all the DC operating points, but they are correct, i.e. like in the simulation. What would you try next?

Thank you very much for your help!

• First question is why on earth you expect such low noise from a JFET at such low frequency and such low drain current. BF862 datasheet shows 0.8 nV/rtHz at a frequency 3 orders of magnitude further away from the problematic 1/F noise region. What in the datasheets or app notes justifies higher expectations? – user_1818839 Feb 3 '19 at 13:47
• Look for "Process 58" FETs in here for a lower noise one at audio frequencies. The data should give an idea how its noise voltage varies with frequency and drain current. google.com/… – user_1818839 Feb 3 '19 at 14:08
• That's correct, in the datasheet the 0.8 nV/rtHz are given for 100 kHz. However, for example on this site you can see that the BF862's voltage noise is at 1.1 nV/rtHz at 1 kHz. This also agrees with the noise from the LTspice simulation. – Charly Feb 3 '19 at 14:28
• So I don't think that it's a matter of the transistor's voltage noise. Is it possible that the gate leakage current is higher in reality, and thus the current noise which transforms to the output via |Zf|? – Charly Feb 3 '19 at 14:35
• Not sure, but your additional evidence looks fairly good for the BF826 (at Id = 1 mA, increasing up to 10mA may help). One area where your amp diverges from tradition (and one of your references) is that the FET is usually a source follower, providing power gain (but no voltage gain). I've used this in a capacitor microphone. TIt keeps impedances (and noise impedances) low ahead of the voltage amp. The other reference DOES show a common source amp - but working into 200R not 10K. That makes me wonder if the trad source follower is worth exploring. – user_1818839 Feb 3 '19 at 14:58

• +1. And also I underline what you stated at the end of your nice answer: as a charge amplifier, a MOSFET input OPAMP with a bias current in the sub-$\mathrm{pA}$ range performs better than almost all JFET input amplifiers, since in this application, what really matters is the input noise current $i_n$ which is directly transferred at the output via a (usually extremely) large feedback resistor. – Daniele Tampieri Apr 4 '19 at 17:49